Remove directory from disposition header (#3515)

PBENCH-1240

Now that the dashboard supports download of tarball inventory, I observed that
the generated filename included a dataset-name prefix. This is because the
content disposition header allows only a name, not a path, and the browser is
trying to be "helpful" by converting the `<dataset-name>/<filename>` string,
replacing the disallowed `/` with `_`. The result is ugly and unexpected.

NOTE: although the Mozilla guidelines try to suggest that the filename must be
enclosed in quotes, this conflicts with the RFC, and Werkzeug's `send_file` is
"reluctant" to include quotes (plus it's ugly to construct). Therefore, this
doesn't attempt to insert the (presumably unnecessary) quotes. (And note also
that, despite the ugly name, the current TOC download works without a quoted
filename in Chrome.)

Instead, report only the name of the file. Add unit and functional test
validation.

lib/pbench/server/api/resources/datasets_inventory.py

@@ -1,4 +1,5 @@
 from http import HTTPStatus
+from pathlib import Path
 from urllib.request import Request
 
 from flask import current_app, send_file
@@ -84,10 +85,9 @@ def _get(
         # We link a callback to close the file stream and TarFile object on
         # completion.
         stream = file_info["stream"]
+        name = Path(file_info["name"]).name
         try:
-            resp = send_file(
-                stream, as_attachment=target is None, download_name=file_info["name"]
-            )
+            resp = send_file(stream, as_attachment=target is None, download_name=name)
         except Exception as e:
             if stream:
                 stream.close()

lib/pbench/test/functional/server/test_datasets.py

@@ -751,6 +751,15 @@ def read_metadata(response: Response) -> JSONOBJECT:
                 API.DATASETS_INVENTORY,
                 {"dataset": dataset.resource_id, "target": "metadata.log"},
             )
+
+            # Note that Werkzeug doesn't understand anything special about the
+            # ".log" filetype, but will apply a better content-type header for
+            # file types it understands like ".json".
+            assert response.headers["content-type"] == "application/octet-stream"
+            assert (
+                response.headers["content-disposition"]
+                == "inline; filename=metadata.log"
+            )
             assert (
                 response.ok
             ), f"INVENTORY {dataset.name} failed {response.status_code}:{response.json()['message']}"

lib/pbench/test/unit/server/test_datasets_inventory.py

@@ -156,3 +156,20 @@ def mock_send_file(path_or_file, *args, **kwargs):
 
         query_get_as("fio_2", "f1.json", HTTPStatus.INTERNAL_SERVER_ERROR)
         assert mock_close
+
+    def test_get_inventory(self, query_get_as, monkeypatch):
+        exp_stream = io.BytesIO(b"file_as_a_byte_stream")
+
+        def mock_get_inventory(_s, _d: str, _t: str):
+            return {
+                "name": "f1.json",
+                "type": CacheType.FILE,
+                "stream": Inventory(exp_stream, None),
+            }
+
+        monkeypatch.setattr(CacheManager, "get_inventory", mock_get_inventory)
+        response = query_get_as("fio_2", "f1.json", HTTPStatus.OK)
+        assert response.status_code == HTTPStatus.OK
+        assert response.text == "file_as_a_byte_stream"
+        assert response.headers["content-type"] == "application/json"
+        assert response.headers["content-disposition"] == "inline; filename=f1.json"